arXiv cs.CL· Reza Habibi, Darian Lee, Magy Seif El-Nasr·· 2 天前AI 评分29
SMDA：将训练数据归因到模型学到的符号化行为策略
Symbolic Mechanistic Data Attribution: Tracing Training Influence to Learned Behavioral Policies
AI 导读
研究者提出符号机制数据归因（SMDA）框架，通过对稀疏自编码器（SAE）特征拟合闭式 Ridge 回归，将每条 SFT 样本的影响分解为特征激活（ΔX）与输出概率（ΔY）两条路径。
正文 · 原文
Abstract:While existing data attribution methods can identify which training examples build specific mechanistic circuits, they cannot explain how training data shapes the behavioral decisions a model learns. We introduce Symbolic Mechanistic Data Attribution (SMDA), a framework that attributes training pairs to the interpretable symbolic policies governing model behavior. SMDA fits a closed-form Ridge regression over sparse autoencoder (SAE) features to model a behavior, then decomposes how each supervised fine-tuning (SFT) example shifts that policy through feature-activation ($\Delta X$) and output-probability ($\Delta Y$) pathways. We distill a symbolic policy for refusal in Llama-3.2-3B-Instruct and analyze 200 SFT pairs. Our analysis reveals (1)~the symbolic policy's coefficients expose systematic gaps in the base model's safety behavior, such as religious stereotyping; (2)~per-feature $\Delta X/\Delta Y$ decomposition mechanistically explains why harmful and harmless pairs exert qualitatively different influences on certain features; and (3)~individual pairs routinely exhibit cross-feature interference, allowing SMDA to identify pairs whose dominant effect falls on unintended features. Retraining without such pairs improved accuracy by up to 4.5 percentage points over full fine-tuning. These results show that combining mechanistic interpretability with data attribution yields a diagnostic tool both more fine-grained than black-box influence functions and more scalable than manual circuit analysis.
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2606.29171 [cs.LG] |
| (or arXiv:2606.29171v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2606.29171 arXiv-issued DOI via DataCite |
Submission history
From: Reza Habibi [view email]
[v1]
Sun, 28 Jun 2026 03:32:06 UTC (1,382 KB)
[v2]
Thu, 15 Oct 2026 17:36:33 UTC (304 KB)
来源：arXiv cs.CL · arxiv.org